A parallel processing approach for implementing high-performance first-order logic deduction systems

Recent trends toward exploiting parallel processing for increasing the execution speed have resulted in the development of parallel algorithms for a variety of applications. The research reported here is devoted to the use of parallel processing for developing high-performance deduction systems, an application in the area of artificial intelligence. During our research we investigated the scheme for parallelizing propositional deduction systems proposed by Butler and Karonis. Specifically, a parallel first-order logic deduction system POTTER-I was developed by applying the scheme of Butler and Karonis to the sequential first-order logic theorem prover OTTER. The performance of POTTER-I was further improved by introducing several optimizations, including message buffering, blocking send, and an encoding scheme for the clauses in the set of support, into the original scheme. In this document, we describe the implementation details of POTTER-I and the enhancements to the original scheme of Butler and Karonis. An evaluation of the enhanced algorithm of POTTER-I for some benchmarks is also presented. 6 figs., 1 tab.